Compare · ModelsLive · 3 picked · head to head
Claude Sonnet 4 vs DeepSeek V3.2 Exp vs Gemini 2.5 Flash
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Claude Sonnet 4 wins on 7/19 benchmarks
Claude Sonnet 4 wins 7 of 19 shared benchmarks. Leads in reasoning.
Category leads
coding·DeepSeek V3.2 Expgeneral·DeepSeek V3.2 Expknowledge·Gemini 2.5 Flashagentic·DeepSeek V3.2 Expreasoning·Claude Sonnet 4arena·DeepSeek V3.2 Expmath·Gemini 2.5 Flash
Hype vs Reality
Attention vs performance
Claude Sonnet 4
#157 by perf·#10 by attention
DeepSeek V3.2 Exp
#133 by perf·no signal
Gemini 2.5 Flash
#191 by perf·no signal
Best value
DeepSeek V3.2 Exp
5.1x better value than Gemini 2.5 Flash
Claude Sonnet 4
5.1 pts/$
$9.00/M
DeepSeek V3.2 Exp
142.4 pts/$
$0.34/M
Gemini 2.5 Flash
28.1 pts/$
$1.40/M
Vendor risk
Mixed exposure
One or more vendors flagged
Anthropic
$965.0B·Tier 1
DeepSeek
$3.4B·Tier 1
Google DeepMind
$4.20T·Tier 1
Head to head
19 benchmarks · 3 models
Claude Sonnet 4DeepSeek V3.2 ExpGemini 2.5 Flash
Aider polyglot
DeepSeek V3.2 Exp leads by +12.9
Aider Polyglot · measures how well AI models can edit code across multiple programming languages using the Aider coding assistant framework.
Claude Sonnet 4
61.3
DeepSeek V3.2 Exp
74.2
Gemini 2.5 Flash
47.1
Dtbench
DeepSeek V3.2 Exp leads by +17.8
Claude Sonnet 4
61.8
DeepSeek V3.2 Exp
79.5
Gemini 2.5 Flash
60.9
Fiction.LiveBench
DeepSeek V3.2 Exp leads by +36.1
Fiction.LiveBench · a continuously updated benchmark using recently published fiction to test reading comprehension and reasoning, preventing data contamination.
Claude Sonnet 4
46.9
DeepSeek V3.2 Exp
83.3
Gemini 2.5 Flash
47.2
Lmca
DeepSeek V3.2 Exp leads by +0.1
Claude Sonnet 4
34.1
DeepSeek V3.2 Exp
34.3
Gemini 2.5 Flash
32.3
The Agent Company
DeepSeek V3.2 Exp leads by +1.8
The Agent Company · tests AI agents on realistic corporate tasks like email management, code review, data analysis, and cross-tool workflows.
Claude Sonnet 4
33.1
DeepSeek V3.2 Exp
42.9
Gemini 2.5 Flash
41.1
WeirdML
Claude Sonnet 4 leads by +5.2
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
Claude Sonnet 4
46.1
DeepSeek V3.2 Exp
39.5
Gemini 2.5 Flash
41.0
APEX-Agents
Claude Sonnet 4 leads by +7.5
APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments.
Claude Sonnet 4
9.3
Gemini 2.5 Flash
1.8
ARC-AGI
Claude Sonnet 4 leads by +7.7
ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization.
Claude Sonnet 4
40.0
Gemini 2.5 Flash
32.3
ARC-AGI-2
Claude Sonnet 4 leads by +3.4
ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data.
Claude Sonnet 4
5.9
Gemini 2.5 Flash
2.5
Chatbot Arena Elo · Overall
DeepSeek V3.2 Exp leads by +12.7
DeepSeek V3.2 Exp
1422.1
Gemini 2.5 Flash
1409.4
DeepResearch Bench
Claude Sonnet 4 leads by +18.6
DeepResearch Bench · evaluates AI on complex multi-step research tasks requiring information gathering, synthesis, and producing comprehensive analyses.
Claude Sonnet 4
47.8
Gemini 2.5 Flash
29.2
FrontierMath-2025-02-28-Private
Gemini 2.5 Flash leads by +1.2
FrontierMath (Feb 2025) · original research-level math problems created by mathematicians, testing capabilities at the boundary of current AI mathematical reasoning.
Claude Sonnet 4
7.3
Gemini 2.5 Flash
8.5
FrontierMath-Tier-4-2025-07-01-Private
Gemini 2.5 Flash leads by +6.9
FrontierMath Tier 4 (Jul 2025) · the most challenging tier of frontier mathematics, containing problems that push the absolute limits of AI mathematical reasoning.
Claude Sonnet 4
0.0
Gemini 2.5 Flash
6.9
GeoBench
Gemini 2.5 Flash leads by +36.0
GeoBench · tests geographic knowledge and spatial reasoning across countries, landmarks, coordinates, and geopolitical understanding.
Claude Sonnet 4
37.0
Gemini 2.5 Flash
73.0
HLE
Gemini 2.5 Flash leads by +4.5
HLE (Humanity's Last Exam) · a reasoning benchmark designed to be the hardest public evaluation of AI. Questions span mathematics, physics, philosophy, and logic · curated to be at or beyond the frontier of human expert capability. Tested with and without tool augmentation. Claude Opus 4.7 scores 46.9% without tools and 54.7% with tools · making it one of the few benchmarks where the top score is below 60%.
Claude Sonnet 4
3.1
Gemini 2.5 Flash
7.7
Lech Mazur Writing
Claude Sonnet 4 leads by +4.9
Lech Mazur Writing · evaluates creative writing ability, assessing prose quality, narrative coherence, and stylistic sophistication.
Claude Sonnet 4
81.4
Gemini 2.5 Flash
76.5
OTIS Mock AIME 2024-2025
Gemini 2.5 Flash leads by +2.0
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
Claude Sonnet 4
71.1
Gemini 2.5 Flash
73.0
SimpleBench
Claude Sonnet 4 leads by +5.2
SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking.
Claude Sonnet 4
34.6
Gemini 2.5 Flash
29.4
VPCT
Gemini 2.5 Flash leads by +6.0
VPCT (Visual Pattern Completion Test) · tests visual reasoning and pattern recognition by having models complete visual sequences and transformations.
Claude Sonnet 4
1.0
Gemini 2.5 Flash
7.0
Full benchmark table
| Benchmark | Claude Sonnet 4 | DeepSeek V3.2 Exp | Gemini 2.5 Flash |
|---|---|---|---|
Aider polyglot Aider Polyglot · measures how well AI models can edit code across multiple programming languages using the Aider coding assistant framework. | 61.3 | 74.2 | 47.1 |
Dtbench | 61.8 | 79.5 | 60.9 |
Fiction.LiveBench Fiction.LiveBench · a continuously updated benchmark using recently published fiction to test reading comprehension and reasoning, preventing data contamination. | 46.9 | 83.3 | 47.2 |
Lmca | 34.1 | 34.3 | 32.3 |
The Agent Company The Agent Company · tests AI agents on realistic corporate tasks like email management, code review, data analysis, and cross-tool workflows. | 33.1 | 42.9 | 41.1 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 46.1 | 39.5 | 41.0 |
APEX-Agents APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments. | 9.3 | — | 1.8 |
ARC-AGI ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization. | 40.0 | — | 32.3 |
ARC-AGI-2 ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data. | 5.9 | — | 2.5 |
Chatbot Arena Elo · Overall | — | 1422.1 | 1409.4 |
DeepResearch Bench DeepResearch Bench · evaluates AI on complex multi-step research tasks requiring information gathering, synthesis, and producing comprehensive analyses. | 47.8 | — | 29.2 |
FrontierMath-2025-02-28-Private FrontierMath (Feb 2025) · original research-level math problems created by mathematicians, testing capabilities at the boundary of current AI mathematical reasoning. | 7.3 | — | 8.5 |
FrontierMath-Tier-4-2025-07-01-Private FrontierMath Tier 4 (Jul 2025) · the most challenging tier of frontier mathematics, containing problems that push the absolute limits of AI mathematical reasoning. | 0.0 | — | 6.9 |
GeoBench GeoBench · tests geographic knowledge and spatial reasoning across countries, landmarks, coordinates, and geopolitical understanding. | 37.0 | — | 73.0 |
HLE HLE (Humanity's Last Exam) · a reasoning benchmark designed to be the hardest public evaluation of AI. Questions span mathematics, physics, philosophy, and logic · curated to be at or beyond the frontier of human expert capability. Tested with and without tool augmentation. Claude Opus 4.7 scores 46.9% without tools and 54.7% with tools · making it one of the few benchmarks where the top score is below 60%. | 3.1 | — | 7.7 |
Lech Mazur Writing Lech Mazur Writing · evaluates creative writing ability, assessing prose quality, narrative coherence, and stylistic sophistication. | 81.4 | — | 76.5 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 71.1 | — | 73.0 |
SimpleBench SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking. | 34.6 | — | 29.4 |
VPCT VPCT (Visual Pattern Completion Test) · tests visual reasoning and pattern recognition by having models complete visual sequences and transformations. | 1.0 | — | 7.0 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| $3.00 | $15.00 | 200K tokens (~100 books) | $60.00 | |
| $0.27 | $0.41 | 164K tokens (~82 books) | $3.05 | |
| $0.30 | $2.50 | 1.0M tokens (~524 books) | $8.50 |